Human-like Computer Vision

Statistical machine learning is widely used in image classification and typically 1) requires many images to achieve high accuracy and 2) does not provide support for reasoning below the level of classification. By contrast this paper describes an approach called machine learning approach called Logical Vision (LV) which uses a) background knowledge such as light reflection that can itself be learned and used for resolving visual ambiguities, which cannot be easily modeled using statistical approaches, b) a wider class of background models representing classical 2D shapes such as circles and ellipses, c) primitive-level statistical estimators to handle noise in real images, Our results indicate that in real images the new noise-robust version of LV using a single example (ie one-shot LV) converges to an accuracy at least comparable to thirty-shot statistical machine learner on the prediction of hidden light sources.